A transformer district cooperative control method and system and a storage medium

Abstract

This invention discloses a method, system, and storage medium for coordinated control of distribution transformer areas, relating to the field of Internet of Things (IoT) technology. The method includes: a distribution transformer control center using a first input convex neural network to output a coordinated control instruction set based on the global state information of the distribution transformer power grid, and distributing a subset of instructions from the coordinated control instruction set to the corresponding edge device control centers; the edge device control centers using a second input convex neural network to output device control instructions based on the received instruction subsets, and sending the device control instructions to the corresponding controlled devices; determining the power grid safety margin based on the first local power grid state information associated with the current edge device control center, and controlling the corresponding controlled devices according to the power grid safety margin. The technical solution of this invention achieves efficient coordination, rapid response, and safe and economical operation of diverse resources within a distribution transformer area.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) technology, and in particular to a method, system, and storage medium for coordinated control of distribution stations. Background Technology

[0002] With the deepening of the global energy transition strategy and the construction of new power systems, the penetration rate of diversified energy sources and storage resources, such as distributed photovoltaics, energy storage systems, and electric vehicle charging piles, in distribution substations is showing an increasing trend. This trend has profoundly changed the operating mode of traditional distribution substations, transforming them from single, passive power distribution terminals into active distribution network nodes with complex interaction characteristics between energy sources, grids, loads, and storage. This transformation brings opportunities to improve energy utilization efficiency and promote the consumption of new energy sources, but it also poses severe challenges to the operation and control technology of distribution substations.

[0003] Currently, existing technical solutions for the coordinated control of diverse resources in a distribution area have the following main shortcomings: First, in terms of response mechanisms, most control strategies rely on centralized computing and command issuance from the master station. Faced with a massive number of distributed access points and rapidly fluctuating operating conditions, communication latency is high and control cycles are long, making it difficult to meet the real-time response requirements for system dynamic stability. Moreover, in terms of setting safety boundaries, existing methods mostly adopt single voltage or power over-limit protection, which easily leads to control strategies that are "one-sided and incomplete." Summary of the Invention

[0004] This invention provides a method, system, and storage medium for coordinated control of distribution transformer areas, in order to solve the problem of coordinated control of diverse resources in distribution transformer areas.

[0005] In a first aspect, the present invention provides a transformer area collaborative control method, applied to a multi-resource collaborative control system, wherein the multi-resource collaborative control system includes a transformer area control center and an edge device control center, wherein:

[0006] The distribution area control center uses a first input convex neural network to output a set of coordinated control instructions based on the global state information of the distribution area power grid, and distributes a subset of instructions in the set of coordinated control instructions to the corresponding edge device control center. The coordinated control instructions in the subset of instructions are control parameters for each type of controlled device.

[0007] The edge device control center uses a second input convex neural network to output device control commands based on a subset of received commands, and sends the device control commands to the corresponding controlled devices; it determines the power grid safety margin based on the first local power grid status information associated with the current edge device control center, and controls the corresponding controlled devices based on the power grid safety margin, wherein the power grid controlled by the distribution area control center includes the local power grid.

[0008] Secondly, the present invention provides a transformer area collaborative control system, which includes a transformer area control center and an edge device control center, wherein:

[0009] The instruction distribution module in the distribution area control center is used to output a set of collaborative control instructions based on the global status information of the distribution area power grid using a first input convex neural network, and distribute a subset of instructions in the collaborative control instruction set to the corresponding edge device control center. The collaborative control instructions in the instruction subset are control parameters for each type of controlled device.

[0010] The control module in the edge device control center is used to output device control commands based on a subset of received commands using a second input convex neural network, and send the device control commands to the corresponding controlled devices; determine the power grid safety margin based on the first local power grid status information associated with the current edge device control center, and control the corresponding controlled devices based on the power grid safety margin, wherein the power grid controlled by the distribution area control center includes the local power grid.

[0011] Thirdly, the present invention provides a computer-readable storage medium storing computer instructions for causing a processor to execute the aforementioned district cooperative control method of the first aspect.

[0012] The transformer area collaborative control scheme provided by this invention:

[0013] 1) A two-tiered architecture was constructed, integrating the distribution area control center and the edge device control center. After generating a collaborative control command set based on global status information, the distribution area control center distributes a subset of commands to the nearest edge device control center, which then generates and executes the device-level control commands locally. This architecture optimizes centralized long-link control into a collaborative mode of "global coordination - local autonomy," significantly shortening the transmission path and execution delay of control commands. This enables the system to quickly respond to instantaneous fluctuations at distributed energy access points, significantly improving the dynamic stability of the distribution area power grid.

[0014] 2) By using an input convex neural network, the complex optimization problem is transformed into a convex optimization form for solution. This enables the control center of the distribution area to quickly output the globally optimal collaborative control instruction set based on the global state information, while the edge device control center can generate accurate device-level control parameters based on the received instruction subset. This achieves end-to-end optimization decision-making from global collaboration to local execution, effectively avoiding the defect of traditional iterative algorithms that are prone to getting trapped in local optima.

[0015] 3) While executing a subset of instructions, the edge device control center of this scheme monitors the status information of the first local power grid in real time, calculates the current safety margin of the power grid, and dynamically constrains and adjusts the device control instructions based on this margin. This mechanism enables the control strategy to maximize the absorption capacity of distributed resources while ensuring the safe operation of the power grid, achieving a coordinated unity between safety and economic objectives, and solving the problems of "over-control" or "under-control" caused by a single over-limit protection mechanism.

[0016] 4) The coordinated control instruction set output by the distribution area control center in this scheme includes control parameters for each type of controlled equipment, achieving fine-grained decomposition of instructions. After receiving the instruction subset, the edge device control center does not mechanically execute it, but rather adaptively adjusts it based on the local power grid status and safety margin. This "global instruction + local correction" mechanism enables the edge device control center to maintain the safe operation of equipment within its area based on local status information even when facing communication interruptions, local equipment failures, or sudden disturbances, significantly improving the system's fault tolerance and operational robustness.

[0017] It should be understood that the description in this section is not intended to identify key or essential features of the invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a transformer area collaborative control method provided in Embodiment 1 of the present invention;

[0020] Figure 2 This is a flowchart of a transformer area collaborative control method provided in Embodiment 2 of the present invention;

[0021] Figure 3 This is a voltage control effect comparison diagram provided by Embodiment 2 of the present invention;

[0022] Figure 4 This is a comparison chart of the power quality improvement effect in a transformer substation according to Embodiment 2 of the present invention;

[0023] Figure 5 This is a schematic diagram of a transformer area collaborative control system according to Embodiment 3 of the present invention;

[0024] Figure 6 This is a schematic diagram of the structure of an electronic device provided according to Embodiment 4 of the present invention. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. In the description of this invention, unless otherwise stated, "a plurality of" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist; for example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0027] Example 1

[0028] Figure 1 This is a flowchart of a transformer area collaborative control method provided in Embodiment 1 of the present invention. This embodiment is applicable to the situation of collaborative control of resources in a transformer area. The method can be executed by a transformer area collaborative control system, which can be implemented in hardware and / or software. The transformer area collaborative control system can be configured in an electronic device, which can be composed of two or more physical entities.

[0029] like Figure 1 As shown, Embodiment 1 of the present invention provides a transformer area collaborative control method, which is applied to a multi-resource collaborative control system. This multi-resource collaborative control system includes a transformer area control center and an edge device control center, and specifically includes the following steps:

[0030] S101. The distribution area control center uses the first input convex neural network to output a set of coordinated control instructions based on the global state information of the distribution area power grid, and distributes the instruction subsets in the set of coordinated control instructions to the corresponding edge device control centers. The coordinated control instructions in the instruction subsets are control parameters for each type of controlled device.

[0031] In this embodiment, a multi-resource collaborative control system with a hierarchical distributed input convex neural network architecture can be pre-constructed. This system includes a distribution area control center deployed with a first input convex neural network and multiple edge device control centers deployed with second input convex neural networks. The first input convex neural network is used to perceive the global state and generate macroscopic collaborative instructions (i.e., collaborative control instructions). The second input convex neural network is used to receive the collaborative instructions and generate the final device control instructions.

[0032] S102. The edge device control center uses the second input convex neural network to output device control commands based on the received instruction subset, and sends the device control commands to the corresponding controlled devices; it determines the power grid safety margin based on the first local power grid status information associated with the current edge device control center, and controls the corresponding controlled devices based on the power grid safety margin, wherein the power grid controlled by the distribution area control center includes the local power grid.

[0033] In this embodiment, the second input convex neural network deployed in the edge device control center can output device control commands based on a subset of received instructions. The edge device control center then sends these commands to the corresponding controlled devices. The edge device control center can also acquire (first) local power grid status information in real time, i.e., the status information of the power grid currently managed by the edge device control center. This information can be used to calculate the power grid safety margin, which includes various power grid indicators. The edge device control center can then control the corresponding controlled devices based on the changing trend of the power grid safety margin.

[0034] The technical solution of this invention, through the technical path of "dual-level input convex neural network architecture + power grid safety margin constraint", solves the technical problems of high communication delay and poor real-time performance in traditional centralized control. At the same time, it overcomes the problem of control strategy incoordination caused by single safety boundary setting, and realizes efficient coordination, rapid response and safe and economical operation of multiple resources in the distribution area.

[0035] Optionally, the distribution area control center uses a first input convex neural network to output a collaborative control instruction set based on the global state information of the distribution area power grid. This includes: the distribution area control center inputs the topology state information of the distribution area power grid into a graph neural network to obtain network connection relationship features, and inputs the network connection relationship features and key information from the global state information into the first input convex neural network to obtain a collaborative control instruction set. The key information is determined based on the controlled devices controlled by the edge device control center, and the collaborative control instruction set includes multiple instruction subsets. The distribution area control center distributes the instruction subsets of the collaborative control instruction set to the corresponding edge device control centers. This includes: the distribution area control center classifies the control instructions in the collaborative control instruction set according to the controlled devices associated with the edge device control centers to obtain multiple instruction subsets, and distributes the multiple instruction subsets to the corresponding edge device control centers.

[0036] For example, if the global state information of the distribution area power grid is a 32-dimensional global state vector... The vector includes eight key node voltages, total active load, total reactive load, photovoltaic power, energy storage power, charging pile power, total harmonic distortion rate, three-phase imbalance, frequency, temperature, irradiance, and hourly information.

[0037] First, a two-layer graph neural network can be used to extract features containing network connectivity relationships from the topological state information of the distribution area power grid and the global state information of the distribution area power grid:

[0038] 1) Data preparation: Collect information on the distribution network in the transformer area, including the number of nodes n and the impedance between each node. .

[0039] 2) Construct the topological adjacency matrix A as follows:

[0040]

[0041] in, The impedance between nodes i and j The reference impedance (taken as in this example) ).

[0042] 3) By Construct the degree matrix D (i.e., the diagonal elements are the weighted degrees of the nodes (the sum of connection strengths)). The expression for D is:

[0043]

[0044] 4) Normalize the adjacency matrix to obtain the matrix :

[0045]

[0046] in, It is an n×n identity matrix.

[0047] 5) Graph Neural Network Message Passing (2 layers):

[0048] Level 0 initialization: (from Extracted node feature matrix)

[0049] First-layer graph neural network computation:

[0050]

[0051] in, This indicates that the features of each node are aggregated with their neighbors. Represents a linear transformation. This represents a column vector of all ones, used for biasing. It is an element-wise ReLU function.

[0052] Second-layer graph neural network computation:

[0053]

[0054] 6) Global pooling: This yields a 16-dimensional feature vector representing network connectivity. .

[0055] Then, the key information from the network connection relationship characteristics and global state information is... The vectors are concatenated to form the input vector of the first input convex neural network. :

[0056]

[0057] in, for transpose, for transpose, Depend on The following was selected:

[0058]

[0059] The components are defined as follows:

[0060] 1) Average state of charge of energy storage:

[0061]

[0062] in, It refers to the number of energy storage systems. It is the kth energy storage moment. The state of charge.

[0063] 2) Total power of charging piles:

[0064]

[0065] in, It refers to the number of charging stations. It is the real-time power of the j-th charging pile.

[0066] 3) Power quality indicators:

[0067]

[0068] in, These are the effective values ​​of the voltage and current corresponding to the h-th harmonic, respectively. These are the fundamental voltage and current RMS values, respectively. It is the negative sequence voltage component.

[0069] 4) The power grid exchanges active and reactive power:

[0070]

[0071] Positive values ​​indicate power absorbed from the upstream power grid, while negative values ​​indicate power fed back to the grid.

[0072] 5) Environmental and time variables:

[0073]

[0074] 6) Load timing characteristics:

[0075] Define time window Hours, sampling interval Minutes, then the number of samples in the window .

[0076]

[0077] (7) Reserved Dimensions (Default value 0, for expansion purposes).

[0078] For example, the first input convex neural network can be a 3-layer structure:

[0079]

[0080] in, (Non-negative weight matrix, ensuring the convexity of the output with respect to the input). It is the ReLU activation function. These are the weighting coefficients. This is the bias coefficient, with no special constraints.

[0081] The first input convex neural network can output multi-dimensional vectors. This refers to the collaborative control instruction set, which can cover the optimized setpoints and key control parameters of resources such as photovoltaics, energy storage, loads, charging piles, and filters.

[0082] Finally, the control center will classify the control commands in the collaborative control command set according to the controlled devices managed by the edge device control center, and distribute the resulting command subsets to the corresponding edge device control centers.

[0083] Optionally, the edge device control center uses a second input convex neural network to output device control commands based on a subset of received commands, including: the edge device control center inputs the received subset of commands and the second local power grid status information associated with the current edge device control center into the second input convex neural network to obtain device control commands.

[0084] Specifically, the input of each second input convex neural network includes (second) local power grid state information (such as local voltage, local power and light intensity) and a subset of received instructions, and the output is device-level fine-grained device control instructions (such as local active or reactive power setpoints, local charging and discharging power and local load regulation limits).

[0085] For example, the general expression for device control instructions is:

[0086]

[0087] in: These are the personalized model parameters for the second input convex neural network obtained through training. It can provide local real-time and high-frequency measurement data (such as local voltage, current, power, state of charge, temperature, and illumination), which can reflect the instantaneous status of the controlled equipment. Different edge device control centers may manage different controlled equipment and receive different subsets of instructions.

[0088] For example, the subset received by the control center of an edge device configured with a photovoltaic inverter may include: total active power setpoint, total reactive power setpoint, voltage reference offset, droop factor, and emergency flag. The subset received by the control center of an edge device configured with an energy storage system may include: charge / discharge power commands, reactive power output commands, operating modes, target state of charge (SOC), voltage reference offset, and emergency flag. The subset received by the control center of an edge device configured with a charging pile may include: total power limit and control modes (such as free charging, orderly charging, and emergency power curtailment). The subset received by the control center of an edge device configured with an active power filter may include: compensation capacity and main control harmonic order. The subset received by the control center of an edge device configured with an adjustable load connection may include: load regulation amount, priority, and emergency flag.

[0089] Optionally, before the edge device control center outputs device control commands based on the received instruction subset using the second input convex neural network, the method further includes: if the edge device control center does not receive the instruction subset issued by the substation control center within a second preset time period, it predicts the instruction subset based on historical instruction subsets using an autoregressive integral moving average model; wherein, the edge device control center outputting device control commands based on the received instruction subset using the second input convex neural network includes: the edge device control center outputting device control commands based on the instruction subset using the second input convex neural network.

[0090] Specifically, during online control, autonomous communication interruption protection is also required. When any edge device control center detects a communication interruption with the area control center lasting longer than a preset duration, it will automatically switch to local autonomous mode. This involves predicting future instruction subsets using historical instruction subsets. The Autoregressive Integrated Moving Average (ARIMA) model can be used to predict instruction subsets based on historical instruction subsets.

[0091]

[0092] in, For the predicted subset of instructions, These are the coefficients of the ARIMA model.

[0093] Optionally, the loss function of the first input convex neural network is determined as follows: the distribution area control center determines the grid economic factor using the real-time electricity price, purchased and sold electricity volume, and active power loss of the distribution area grid; determines the voltage quality factor based on the real-time node voltage and preset voltage deviation threshold of the distribution area grid; determines the power quality factor based on the real-time voltage harmonic distortion rate and real-time three-phase voltage imbalance of the distribution area grid; and determines the safety constraint violation degree factor based on the target parameters of the distribution area grid, wherein the target parameters include real-time node voltage, real-time branch power, real-time branch current, real-time grid frequency, real-time apparent power of equipment, real-time state of charge of energy storage, and power change rate; and the grid economic factor, the voltage quality factor, the power quality factor, and the safety constraint violation degree factor are used to determine the power grid economic factor, the voltage quality factor, the power quality factor, and the safety constraint violation degree factor. The loss function of the first input convex neural network is obtained by weighted summation. The loss function of the second input convex neural network is determined by: the edge device control center determining the first squared value of the Euclidean distance between the device control command output by the second input convex neural network during training and the input quantity, wherein the input quantity is a subset of the commands output by the trained first input convex neural network; determining the second squared value of the Euclidean distance between the device control command output by the second input convex neural network during training and the local optimal command, wherein the local optimal command is a control command obtained after processing historical local power grid states based on preset rules; and weighted summation of the first squared value and the second squared value to obtain the loss function of the second input convex neural network.

[0094] Specifically, a two-stage collaborative training of the first and second input convex neural networks can address the convergence difficulties and poor performance issues of direct distributed training. The first stage (centralized pre-training) involves training the first input convex neural network using a training dataset, enabling it to learn the globally optimal collaborative control law under multiple safety constraints. The second stage (distributed fine-tuning) uses partial network weights obtained from the training of the first input convex neural network to initialize each second input convex neural network. Then, distributed transfer learning is performed on the initialized second input convex neural networks using local operating data from each controllable resource (the training dataset for the second input convex neural networks), adapting them to the physical characteristics and control objectives of the local resources.

[0095] in:

[0096] Phase 1: Intensive training of the first input convex neural network:

[0097] Objective: To enable the first input convex neural network to learn the globally optimal cooperative control law under multiple security constraints.

[0098] Specifically, this can be understood as using a training dataset to allow the first input convex neural network to learn the global collaborative optimization rules under multiple constraints such as safety, economy, and quality.

[0099] Loss Function Design: Design a multi-objective integrated loss function It also optimizes economic efficiency, voltage quality, and power quality, and penalizes violations of safety constraints.

[0100]

[0101] in:

[0102] , maximum.

[0103] Power grid economic factors , Let be the unit cost of power exchanged with the grid at time t. Let be the active power exchanged with the upstream power grid at time t. This is the network loss cost coefficient. Let Δt be the total active power loss within the distribution area at time t, and Δt be the time step. This reflects the cost of purchasing electricity or the revenue from selling electricity (negative cost). This reflects the economic losses caused by network failures.

[0104] Voltage quality factor , Let be the voltage magnitude of node i at time t. This is the voltage reference value. This is the preset maximum allowable voltage deviation. `max(0,·)` indicates that only the portion exceeding the allowable range is penalized; within the allowable range... The loss is 0 when ().

[0105] Power quality factor THD(t) is the total harmonic distortion of the voltage at time t. Let t be the three-phase voltage imbalance at time t.

[0106] Safety constraint violation factor constraint_violation is a comprehensive quantification of the degree of violation of various security constraints. m is the number of constraints. This represents the weight coefficient for the m-th constraint. (Safety constraint violation item) include:

[0107] Voltage safety: determined based on the value by which the node voltage exceeds a preset hard safety boundary;

[0108] Branch power safety: determined based on the value by which the active power of the branch exceeds the preset thermal stability limit;

[0109] Branch current safety: determined based on the value of the branch current exceeding the preset thermal stability limit;

[0110] Frequency safety: determined based on the value at which the frequency exceeds the preset allowable range;

[0111] Equipment capacity safety: determined based on the amount by which the apparent power of the equipment exceeds the rated capacity;

[0112] Energy storage SOC safety: determined based on the value of SOC exceeding the preset allowable range;

[0113] Gradeability safety: The value at which the power change rate exceeds the equipment limit is determined.

[0114] The coefficients before the safety constraint violation are different. They can be different from each other.

[0115] Training parameters: Adam optimizer, learning rate 0.001, batch size 256, training for 200 epochs.

[0116] Training Results: After training, the first input convex neural network is capable of generating high-quality macroscopic cooperative instructions. The ability.

[0117] Phase Two: Distributed Fine-tuning of the Second Input Convex Neural Network

[0118] Objective: Based on the global knowledge already possessed by the first input convex neural network, enable each second input convex neural network to quickly adapt to the local physical characteristics and control objectives of its corresponding resources.

[0119] Method: Transfer learning is employed, using the weights of the resource-related network components in the first input convex neural network as the initial values ​​for the corresponding second input convex neural network. Then, each edge node utilizes its local historical data, guided by collaborative control commands and with local optimality as the reference target, to apply the loss function of the second input convex neural network. The model is then fine-tuned. This stage achieves accurate local adaptation while maintaining the global optimization objective.

[0120] Transfer learning initialization: The weights related to each resource in the first input convex neural network are used as the initial parameters of the corresponding second input convex neural network. .

[0121]

[0122] In the formula, The model parameters are obtained by training the convex neural network with the first input.

[0123] Local data fine-tuning: Each edge node uses its local dataset (such as one month of historical data) to fine-tune the model. Loss function Balancing the following of the second input convex neural network with the approximation of the local optimum:

[0124]

[0125] In the formula, This is the regularization coefficient (which can be 0.5). It is the square of the Euclidean distance between the device control command and the cooperative control command, which can constrain the edge behavior to not deviate from the global optimization intention of the center. This refers to the local optimal instruction based solely on local state and target, without considering global coordination.

[0126] Determination method: Taking expert rule-based method as an example:

[0127]

[0128] Rule_local() is an expert rule function.

[0129] The second input convex neural network can be trained using an exponentially decaying learning rate (e.g., an initial learning rate of 0.001) and an early stopping mechanism (e.g., stopping if the validation loss does not decrease after 10 rounds).

[0130] Finally, the trained input convex neural networks can be deployed to the corresponding hardware units to achieve online hybrid control. The first input convex neural network runs at a first preset period (e.g., seconds), outputting and issuing a set of coordinated control instructions based on the global state information of the power grid in the distribution area. Each second input convex neural network runs at a second preset period (e.g., milliseconds) shorter than the first preset period, outputting and executing fine-grained device control instructions at the device level based on a subset of the received instructions.

[0131] Optionally, before training the first and second input convex neural networks, historical operating data of the distribution area can be collected in advance and simulation data can be generated, which can then be fused to construct a dataset for model training. Historical operating data of the distribution area (including SCADA, power quality monitoring, and resource output records, etc.) can be collected and combined with power system simulation software to generate simulation data covering typical and extreme scenarios. These two datasets can then be fused to form a high-quality training dataset pool covering all operating conditions. A central training dataset containing the mapping relationship between the global state information of the distribution area power grid and the optimal coordinated control command set, and an edge training dataset containing the mapping relationship between local state, a subset of commands, and the optimal equipment control commands can be constructed. By fusing historical data and simulation data, a high-quality dataset covering all operating conditions can be built.

[0132] For example, the process of constructing a training dataset may include:

[0133] 1. Define data requirements and collection points:

[0134] according to ), clearly define the specific content, source, and frequency of data collection:

[0135] (1) Data collection points in the distribution area (used for training the first input convex neural network) include: the SCADA system of the distribution area collaborative control center, power quality monitoring devices, distribution transformer terminals (TTUs), and distributed power source monitoring systems, etc., used to collect global status information of the distribution area power grid. Time stamps, historical or predicted load and photovoltaic curves can be generated by the background system.

[0136] (2) Edge data acquisition points (for training the second input convex neural network) include: sensors and communication modules built into the local controllers of each controllable resource (such as photovoltaic inverters, energy storage converters, charging pile controllers, active filter controllers and adjustable load controllers, etc.), which can collect local high-frequency data such as voltage, current, power, SOC, temperature and light.

[0137] (3) Control commands and effect records (used to construct training labels): Record the optimization settings issued by the central optimization software (or historical manual scheduling commands) as... The reference label records the actual control commands executed by each resource's local controller (such as the active or reactive power setpoints of the inverter and the charging and discharging power commands of the energy storage, etc.) as... Reference tags.

[0138] 2. Construct the training dataset for the first input convex neural network:

[0139] Objective: To construct a structure like Data pairs, among which It corresponds to The optimal (or suboptimal feasible) cooperative control command in the given state.

[0140] Historical data extraction and preprocessing:

[0141] Extraction: Extract from historical databases (SCADA, monitoring systems) by timestamp alignment. Historical data for all dimensions required in the process.

[0142] Cleaning: Handling missing values ​​(e.g., interpolation) and outliers (e.g., removal or correction based on statistical methods).

[0143] Note: For each historical moment Calculate or match a "cooperative control instruction" as a training label. This is typically obtained through the following methods:

[0144] Method 1 (based on historical optimization records): If there are historical records of instructions issued by the central optimization platform, use them directly. .

[0145] Method 2 (based on ex-post optimal inversion): using historical moments Starting from the initial state, with objectives such as economy, safety, and power quality, run an offline optimization calculation (such as using traditional optimization algorithms like mixed integer programming or model predictive control), and use the optimization result as... .

[0146] Method 3 (based on expert rules and safety constraints): Based on operating procedures and safety upper and lower limits, for The state generates a feasible set of instructions that satisfies all constraints, from which a representative instruction is selected as... .

[0147] Simulation data generation (supplementation and enhancement):

[0148] Scenario design: Use power system simulation software (such as MATLAB / Simulink) to build detailed models of the transformer substations, including photovoltaics, energy storage, charging piles, adjustable loads, and active power filters.

[0149] Parameter and perturbation settings: simulation Multiple combinations of various variables, especially extreme or boundary scenarios that are scarce in historical data (such as extremely high / low photovoltaic output, heavy overload, severe voltage exceedance, harmonic exceedance, or equipment failure).

[0150] Optimization and command generation: For each simulation scenario, Method 2 (offline optimization calculation) or a combination of a precise physical model and optimization algorithm is used to solve for the optimal cooperative control command that satisfies multiple objectives and constraints in that scenario. .

[0151] Data recording: Records the system state after each simulation scenario has stabilized (as...). ) and its corresponding optimization instructions .

[0152] Data fusion and format unification:

[0153] Fusion: The processed historical data is merged with the generated simulation data to form a central training dataset pool that is comprehensive and rich in operating conditions.

[0154] GNN Feature Pre-extraction (Optional but Efficient): Based on the structure of the input convex neural network, extract the topological adjacency matrix A and initial node features for each sample in the dataset pool. (from (Extracting node-level information) is performed for forward computation to obtain the multidimensional GNN output feature vector corresponding to each sample. Then With the corresponding (from (Selected from the middle) are concatenated to directly generate the final 32-dimensional input vector for each sample. In this way, the first input to the convex neural network during training is directly... .

[0155] Partitioning: Dividing the final dataset ( The dataset is randomly divided into training, validation, and test sets in a ratio of 8:1:1.

[0156] 3. Construct the training dataset for the second input convex neural network:

[0157] Objective: To construct a system for each resource type (such as photovoltaics and energy storage) in the form of... The data pairs.

[0158] Data association and alignment:

[0159] 1) For each resource type, extract from its local historical data (Such as local voltage, current, active power, reactive power, irradiance, local voltage, current, SOC, charging and discharging power of photovoltaics, etc.).

[0160] 2) Based on the timestamp, find the subset of macroscopic instructions related to the resource issued at the same time from the training dataset of the first input convex neural network.

[0161] 3) From the history of the local controller of this resource, find the control instructions that were actually executed at the same time and had good effects. (As a training label). A good result can be judged by whether local control objectives (such as tracking setpoint accuracy or local voltage stability) are achieved.

[0162] Data supplementation and generation:

[0163] 1) In the simulation model, not only can coordinated control commands and global power grid status information of the distribution area be recorded, but also the local measurement values ​​of each resource node can be accurately recorded. and in and received Under certain conditions, control commands generated by a precise local controller model (or local-level optimization) .

[0164] (2) Combine historical data with simulation-generated data to form training datasets for various resources.

[0165] Dataset partitioning:

[0166] For each resource type, the training dataset is divided into training, validation, and test sets in a ratio (e.g., 8:1:1).

[0167] Example 2

[0168] Figure 2 This is a flowchart of a transformer area collaborative control method provided in Embodiment 2 of the present invention. The technical solution of the present invention is further optimized based on the above optional technical solutions, and provides a specific way to collaboratively control resources in a transformer area.

[0169] Optionally, the controlled equipment in the distribution network includes photovoltaics, energy storage, charging piles, active power filters, and adjustable loads. The distribution network control center inputs the distribution network topology status information into a graph neural network to obtain network connection characteristics. It then inputs these network connection characteristics and key information from the global status information into a first input convex neural network to obtain a coordinated control instruction set. This includes: the distribution network control center inputting voltage information, power information, power quality information, environmental and time information, and timing characteristics of the distribution network into the graph neural network to obtain network connection characteristics. The center then inputs these network connection characteristics and key information from the global status information into the first input convex neural network to obtain a coordinated control instruction set. The power information includes at least the power information of photovoltaics, energy storage, and charging piles. The coordinated control instruction set includes photovoltaic control parameters, energy storage control parameters, charging pile control parameters, active power filter control parameters, and adjustable load control parameters.

[0170] Optionally, the edge device control center determines the grid safety margin based on the first local grid status information associated with the current edge device control center, and controls the corresponding controlled device according to the grid safety margin, including: the edge device control center determines the voltage safety margin, load safety margin, and power quality margin respectively based on the real-time node voltage and preset voltage lower limit, real-time branch power and rated power, and real-time voltage harmonic distortion rate and three-phase voltage imbalance in the associated local grid; triggers control warnings based on the voltage safety margin, the load safety margin, and the power quality margin, and controls the corresponding controlled device according to the preset control strategy corresponding to the control warning.

[0171] Optionally, the above method further includes: the distribution area control center determining voltage safety margin, load safety margin, and power quality safety margin respectively based on the real-time node voltage and preset reference voltage, real-time branch power and rated power, real-time voltage harmonic distortion rate, and three-phase voltage imbalance in the associated distribution area power grid; weighted summing of the voltage safety margin, the load safety margin, and the power quality safety margin to obtain the distribution area power grid safety margin; if the distribution area power grid safety margin continues to decrease within a first preset time period, the edge device control center is instructed to increase the acquisition frequency of real-time node voltage, real-time branch power and rated power, and real-time voltage harmonic distortion rate, as well as the determination frequency of the voltage safety margin, the load safety margin, and the power quality margin.

[0172] like Figure 2 As shown, the second embodiment of the present invention provides a method for coordinated control of transformer substations, which specifically includes the following steps:

[0173] S201. The distribution area control center inputs the voltage information, power information, power quality information, environmental and time information, and timing characteristics of the distribution area power grid into a graph neural network to obtain network connection relationship characteristics. The network connection relationship characteristics and key information from the global state information are then input into a first input convex neural network to obtain a set of coordinated control instructions.

[0174] The controlled equipment in the power grid includes photovoltaics, energy storage, charging piles, active power filters, and adjustable loads. The power information includes at least the power information of photovoltaics, energy storage, and charging piles. The coordinated control command set includes photovoltaic control parameters, energy storage control parameters, charging pile control parameters, active power filter control parameters, and adjustable load control parameters.

[0175] S202. The control center of the distribution area classifies the control instructions in the collaborative control instruction set according to the controlled devices associated with the edge device control center to obtain multiple instruction subsets, and distributes the multiple instruction subsets to the corresponding edge device control centers.

[0176] S203. The edge device control center inputs the received instruction subset and the second local power grid status information associated with the current edge device control center into the second input convex neural network to obtain device control instructions, and sends the device control instructions to the corresponding controlled device.

[0177] S204. The edge device control center determines the voltage safety margin, load safety margin, and power quality margin respectively according to the real-time node voltage and preset voltage lower limit, real-time branch power and rated power, and real-time voltage harmonic distortion rate and three-phase voltage unbalance degree in the associated local power grid; triggers control warnings according to the voltage safety margin, the load safety margin, and the power quality margin, and controls the corresponding controlled devices according to the preset control strategies corresponding to the control warnings.

[0178] Exemplarily, the methods for determining the voltage safety margin, load safety margin, and power quality margin include:

[0179] Voltage safety margin :

[0180]

[0181] Load safety margin :

[0182]

[0183] Power quality margin :

[0184]

[0185] Wherein, is the voltage amplitude of node i at time t, is the preset voltage lower limit, is the preset voltage upper limit, is the active power transmitted by branch j at time t, is the rated capacity of branch j. [[ID=4​​​​​​​​​​​​​​​​​​​​​​​% and the voltage is too high ( Reverse power leads to line load factor If so, a Level 3 warning will be triggered.

[0193] S205. The distribution area control center determines the voltage safety margin, load safety margin, and power quality safety margin based on the real-time node voltage and preset reference voltage, real-time branch power and rated power, real-time voltage harmonic distortion rate, and three-phase voltage imbalance in the associated distribution area power grid. The voltage safety margin, load safety margin, and power quality safety margin are weighted and summed to obtain the distribution area power grid safety margin. If the distribution area power grid safety margin continues to decrease within a first preset time period, the edge device control center is instructed to increase the acquisition frequency of real-time node voltage, real-time branch power and rated power, and real-time voltage harmonic distortion rate, as well as the determination frequency of the voltage safety margin, load safety margin, and power quality margin.

[0194] Specifically, the methods for determining the safety level S(t) of the power grid in a distribution area include:

[0195]

[0196] in:

[0197]

[0198] The weighting coefficient can be taken as follows: ,satisfy .

[0199] In the above formula, For the preset reference voltage, The preset three-phase voltage imbalance threshold, This is a preset threshold for voltage harmonic distortion rate, which is a fixed limit (e.g., 4%). Power quality safety is used to measure the distance between the current power quality and the safety boundary: the higher the value, the safer it is.

[0200] When S(t) continues to decrease, even if the voltage safety margin, load safety margin, or power quality margin has not yet triggered an early warning, it can enter the "attention" state in advance, instructing the edge device control center to increase the acquisition frequency of real-time node voltage, real-time branch power and rated power, and real-time voltage harmonic distortion rate, as well as the determination frequency of voltage safety margin, load safety margin, and power quality margin, in order to improve the monitoring frequency.

[0201] Optionally, under heavy load, the line voltage drop increases. If fixed voltage limits are still applied, it can easily lead to discrepancies in judgments such as "voltage is qualified but line is overloaded" or "line is not overloaded but voltage exceeds the limit." The distribution area control center can also set power grid indicator constraint boundaries, such as:

[0202] Voltage-load coupling constraints:

[0203]

[0204] in, It is the voltage lower limit contraction factor, which indicates that when the line is heavily loaded, the voltage drop increases. In order to avoid the voltage being too low, the actual allowable voltage lower limit should be appropriately increased.

[0205] This is the voltage upper limit contraction coefficient, which indicates that when the line is heavily loaded, if the voltage is too high, it may aggravate the heating of the equipment, so the upper limit should be appropriately reduced.

[0206] It is the dynamic coupling coefficient The adaptive adjustment parameter can be taken from 0.5 to 2.0. The reference coupling coefficient in the voltage-load coupling constraint model can be taken as 0.01 to 0.05 pu.

[0207] when If the above constraints are not met, the system will be classified as "alert" or "emergency," thereby triggering control measures in advance, such as energy storage support or photovoltaic power curtailment.

[0208] Voltage-power quality coupling constraints:

[0209]

[0210] in, The sensitivity coefficient ranges from 0.5 to 1.0. This serves as the benchmark value for voltage harmonic distortion rate. This constraint reflects the physical law that when the voltage deviates from the reference value, the nonlinear characteristics of the equipment intensify and the harmonic level rises. The greater the deviation of the voltage from the rated value, the stricter the allowable harmonic limits become, avoiding a double deterioration of both voltage and power quality.

[0211] It will dynamically tighten based on the current voltage. For example, if... =3%, when the voltage is normal ( ), At this point, the permissible upper limit for harmonics is 3%. When the voltage exceeds this limit (e.g., ...), ., , ), At this point, the allowable upper limit for harmonics is relaxed to 9.4%, and the constraint boundary dynamically expands. The system's tolerance for harmonics actually increases. However, precisely because of this increased tolerance, stricter harmonic control is necessary. Once the voltage exceeds the limit, even if the harmonic level remains unchanged, immediate action must be taken; otherwise, the voltage problem may worsen. When the voltage exceeds the limit, voltage safety decreases significantly, amplifying the impact of power quality safety on overall safety and increasing the sensitivity of power quality safety to harmonic changes. Furthermore, when the voltage is too high or too low and harmonics are close to the limit, a higher level of warning can be initiated earlier, prioritizing voltage adjustment (such as reactive power support from energy storage or photovoltaic reactive power voltage regulation) rather than simply addressing harmonics. During training, the second input convex neural network... Includes and These instructions, combined with the current voltage deviation, will proactively tighten the harmonic control target to prevent secondary power quality problems caused by control actions.

[0212] Optionally, feedback data on the actual control effect can be collected, and online incremental learning can be performed on the first input convex neural network and / or the second input convex neural network based on the feedback data to achieve continuous optimization of the model.

[0213] For example, Figure 3 This is a comparison chart of voltage control effects. Figure 4 This is a comparison chart showing the effect of power quality improvement in a transformer substation. The comparison chart showing the voltage control effect in substations where this method was applied is shown below. Figure 3 As shown in the figure, the comparison chart of the power quality improvement effect in the transformer area is as follows: Figure 4 As shown.

[0214] The transformer substation collaborative control method provided in this invention proposes a hierarchical distributed architecture of "transformer substation collaborative control + edge autonomy," realizing a hybrid control of "central second-level collaborative optimization and edge millisecond-level rapid autonomy." This improves the system's effective response speed from the traditional centralized second level to the millisecond level, meeting the stringent requirements of fast control in power-electronic transformer substations. This method incorporates communication interruption autonomy, enabling the edge device control center to continue operating independently even after losing contact with the transformer substation control center, ensuring basic safety. This method establishes a dynamically coupled mathematical model of three safety boundaries: voltage, load, and power quality, achieving collaborative optimization and control of multiple safety objectives. This method supports plug-and-play new resources and continuous adaptive optimization of the system under new operating conditions, demonstrating significant engineering application value.

[0215] Example 3

[0216] Figure 5 This is a schematic diagram of a transformer area collaborative control system provided in Embodiment 3 of the present invention. Figure 5As shown, the system includes: a control center 301 for the distribution area and a control center 302 for the edge devices, wherein:

[0217] The instruction distribution module in the distribution area control center is used to output a set of collaborative control instructions based on the global state information of the distribution area power grid using a first input convex neural network, and distribute a subset of instructions in the set of collaborative control instructions to the corresponding edge device control center. The collaborative control instructions in the subset of instructions are control parameters for each type of controlled device.

[0218] The control module in the edge device control center is used to output device control commands based on a subset of received commands using a second input convex neural network, and send the device control commands to the corresponding controlled devices; determine the power grid safety margin based on the first local power grid status information associated with the current edge device control center, and control the corresponding controlled devices based on the power grid safety margin, wherein the power grid controlled by the distribution area control center includes the local power grid.

[0219] The transformer substation collaborative control system provided in this embodiment of the invention solves the technical problems of high communication delay and poor real-time performance in traditional centralized control by adopting the technical approach of "dual-level input convex neural network architecture + power grid safety margin constraint". At the same time, it overcomes the problem of control strategy incoordination caused by single safety boundary setting, and realizes efficient collaboration, rapid response and safe and economical operation of multiple resources in the transformer substation.

[0220] Optionally, the instruction issuance module includes:

[0221] The instruction generation unit is used to input the power grid topology status information of the distribution area into a graph neural network to obtain network connection relationship features, and input the network connection relationship features and key information in the global status information into a first input convex neural network to obtain a collaborative control instruction set. The key information is determined according to the controlled equipment controlled by the edge device control center, and the collaborative control instruction set includes multiple instruction subsets.

[0222] The distribution unit is used to classify the control instructions in the collaborative control instruction set according to the controlled devices associated with the edge device control center to obtain multiple instruction subsets, and distribute the multiple instruction subsets to the corresponding edge device control centers.

[0223] Furthermore, the controlled equipment in the distribution network includes photovoltaics, energy storage, charging piles, active power filters, and adjustable loads. The distribution network control center inputs the distribution network topology status information into a graph neural network to obtain network connection characteristics. It then inputs these network connection characteristics and key information from the global status information into a first input convex neural network to obtain a collaborative control instruction set. This includes: the distribution network control center inputting voltage information, power information, power quality information, environmental and time information, and timing characteristics of the distribution network into the graph neural network to obtain network connection characteristics. The power information includes at least the power information of photovoltaics, energy storage, and charging piles. The collaborative control instruction set includes photovoltaic control parameters, energy storage control parameters, charging pile control parameters, active power filter control parameters, and adjustable load control parameters.

[0224] Optionally, the control module includes:

[0225] The instruction generation unit is used to input the received instruction subset and the second local power grid status information associated with the current edge device control center into the second input convex neural network to obtain device control instructions.

[0226] Optionally, the control module includes:

[0227] The early warning processing unit is used to determine the voltage safety margin, load safety margin, and power quality margin based on the real-time node voltage and preset voltage lower limit, real-time branch power and rated power, real-time voltage harmonic distortion rate, and three-phase voltage imbalance in the associated local power grid, respectively; trigger control early warning based on the voltage safety margin, the load safety margin, and the power quality margin, and control the corresponding controlled equipment according to the preset control strategy corresponding to the control early warning.

[0228] Optionally, the control module may also include:

[0229] The indicating unit is used to determine voltage safety margin, load safety margin, and power quality safety margin based on the real-time node voltage and preset reference voltage, real-time branch power and rated power, real-time voltage harmonic distortion rate, and three-phase voltage imbalance in the associated distribution network; to obtain the distribution network safety margin by weighted summation of the voltage safety margin, the load safety margin, and the power quality safety margin; if the distribution network safety margin continues to decrease within a first preset time period, the unit instructs the edge device control center to increase the acquisition frequency of real-time node voltage, real-time branch power and rated power, and real-time voltage harmonic distortion rate, as well as to increase the determination frequency of the voltage safety margin, the load safety margin, and the power quality margin.

[0230] Optionally, the system may also include:

[0231] The prediction module in the edge device control center is used to predict the instruction subset based on the historical instruction subset if it does not receive the instruction subset issued by the control center within a second preset time period before outputting the device control instruction based on the received instruction subset using the second input convex neural network.

[0232] The edge device control center uses a second input convex neural network to output device control commands based on a subset of received commands, including: the edge device control center uses the second input convex neural network to output device control commands based on the subset of commands.

[0233] Optionally, the loss function of the first input convex neural network is determined as follows: the distribution area control center determines the grid economic factor using the real-time electricity price, purchased and sold electricity volume, and active power loss of the distribution area grid; determines the voltage quality factor based on the real-time node voltage and preset voltage deviation threshold of the distribution area grid; determines the power quality factor based on the real-time voltage harmonic distortion rate and real-time three-phase voltage imbalance of the distribution area grid; determines the safety constraint violation degree factor based on the target parameters of the distribution area grid, wherein the target parameters include real-time node voltage, real-time branch power, real-time branch current, real-time grid frequency, real-time apparent power of equipment, real-time state of charge of energy storage, and power change rate; and performs a weighted summation of the grid economic factor, the voltage quality factor, the power quality factor, and the safety constraint violation degree factor to obtain the loss function of the first input convex neural network.

[0234] The method for determining the loss function of the second input convex neural network includes: the edge device control center determining the first squared value of the Euclidean distance between the device control command output by the second input convex neural network during training and the input quantity, wherein the input quantity is a subset of the commands output by the first input convex neural network after training; determining the second squared value of the Euclidean distance between the device control command output by the second input convex neural network during training and the local optimal command, wherein the local optimal command is a control command obtained after processing the historical local power grid state based on preset rules; and performing a weighted summation of the first squared value and the second squared value to obtain the loss function of the second input convex neural network.

[0235] The transformer area collaborative control system provided in the embodiments of the present invention can execute the transformer area collaborative control method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0236] Example 4

[0237] Figure 6A schematic diagram of an electronic device 40 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0238] like Figure 6 As shown, the electronic device 40 includes at least one processor 41 and a memory, such as a read-only memory (ROM) 42 or a random access memory (RAM) 43, communicatively connected to the at least one processor 41. The memory stores computer programs executable by the at least one processor. The processor 41 can perform various appropriate actions and processes based on the computer program stored in the ROM 42 or loaded from storage unit 48 into the RAM 43. The RAM 43 may also store various programs and data required for the operation of the electronic device 40. The processor 41, ROM 42, and RAM 43 are interconnected via a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0239] Multiple components in electronic device 40 are connected to I / O interface 45, including: input unit 46, such as keyboard, mouse, etc.; output unit 47, such as various types of monitors, speakers, etc.; storage unit 48, such as disk, optical disk, etc.; and communication unit 49, such as network card, modem, wireless transceiver, etc. Communication unit 49 allows electronic device 40 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0240] Processor 41 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 41 performs the various methods and processes described above, such as the area cooperative control method.

[0241] In some embodiments, the transformer area collaborative control method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 48. In some embodiments, part or all of the computer program may be loaded into and / or installed on electronic device 40 via ROM 42 and / or communication unit 49. When the computer program is loaded into RAM 43 and executed by processor 41, one or more steps of the transformer area collaborative control method described above may be performed. Alternatively, in other embodiments, processor 41 may be configured to perform the transformer area collaborative control method by any other suitable means (e.g., by means of firmware).

[0242] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoC) systems, complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0243] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0244] The computer equipment provided above can be used to execute the transformer area collaborative control method provided in any of the above embodiments, and has corresponding functions and beneficial effects.

[0245] In some embodiments, the transformer area collaborative control method of the present invention may also be stored in a computer-readable storage medium, thereby enabling a processor in a computer to execute the transformer area collaborative control method, the method comprising:

[0246] The distribution area control center uses the first input convex neural network to output a set of coordinated control instructions based on the global state information of the distribution area power grid, and distributes the instruction subsets in the set of coordinated control instructions to the corresponding edge device control centers. The coordinated control instructions in the instruction subsets are control parameters for each type of controlled device.

[0247] The edge device control center uses a second input convex neural network to output device control commands based on a subset of received commands and sends the device control commands to the corresponding controlled devices. It determines the power grid safety margin based on the first local power grid status information associated with the current edge device control center and controls the corresponding controlled devices based on the power grid safety margin. The power grid controlled by the distribution area control center includes the local power grid.

[0248] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by, or in conjunction with, an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0249] The computer equipment provided above can be used to execute the transformer area collaborative control method provided in any of the above embodiments, and has corresponding functions and beneficial effects.

[0250] It is worth noting that in the above embodiments of the coordinated control system for distribution areas, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0251] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A method for coordinated control of transformer substations, characterized in that, This is applied to a multi-resource collaborative control system, which includes a distribution area control center and an edge device control center, wherein: The distribution area control center uses a first input convex neural network to output a set of collaborative control instructions based on the global state information of the distribution area power grid, and distributes a subset of instructions in the set of collaborative control instructions to the corresponding edge device control center. The collaborative control instructions in the subset of instructions are control parameters for each type of controlled device. The edge device control center uses a second input convex neural network to output device control commands based on a subset of received commands, and sends the device control commands to the corresponding controlled devices; it determines the power grid safety margin based on the first local power grid status information associated with the current edge device control center, and controls the corresponding controlled devices based on the power grid safety margin, wherein the power grid controlled by the distribution area control center includes the local power grid.

2. The method according to claim 1, characterized in that, The distribution area control center uses a first input convex neural network to output a set of coordinated control commands based on the global state information of the distribution area power grid, including: The distribution area control center inputs the distribution area power grid topology status information into a graph neural network to obtain network connection relationship features, and inputs the network connection relationship features and key information from the global status information into a first input convex neural network to obtain a collaborative control instruction set. The key information is determined according to the controlled equipment controlled by the edge device control center, and the collaborative control instruction set includes multiple instruction subsets. The control center of the distribution area distributes a subset of instructions from the collaborative control instruction set to the corresponding edge device control center, including: The control center of the distribution area classifies the control instructions in the collaborative control instruction set according to the controlled devices associated with the edge device control center to obtain multiple instruction subsets, and distributes the multiple instruction subsets to the corresponding edge device control centers.

3. The method according to claim 2, characterized in that, The controlled equipment in the distribution network includes photovoltaics, energy storage, charging piles, active power filters, and adjustable loads. The distribution network control center inputs the distribution network topology state information into a graph neural network to obtain network connectivity features. It then inputs these network connectivity features and key information from the global state information into a first input convex neural network to obtain a set of coordinated control instructions, including: The distribution area control center inputs the voltage information, power information, power quality information, environmental and time information, and time-series characteristics of the distribution area power grid into a graph neural network to obtain network connection relationship characteristics. The network connection relationship characteristics and key information from the global state information are then input into a first input convex neural network to obtain a collaborative control instruction set. The power information includes at least the power information of photovoltaic, energy storage, and charging piles. The collaborative control instruction set includes photovoltaic control parameters, energy storage control parameters, charging pile control parameters, active filter control parameters, and adjustable load control parameters.

4. The method according to any one of claims 1-3, characterized in that, The edge device control center uses a second input convex neural network to output device control commands based on a subset of received commands, including: The edge device control center inputs the received instruction subset and the second local power grid status information associated with the current edge device control center into the second input convex neural network to obtain device control instructions.

5. The method according to any one of claims 1-3, characterized in that, The edge device control center determines the power grid security margin based on the first local power grid status information associated with the current edge device control center, and controls the corresponding controlled devices according to the power grid security margin, including: The edge device control center determines the voltage safety margin, load safety margin, and power quality margin based on the real-time node voltage and preset voltage lower limit, real-time branch power and rated power, real-time voltage harmonic distortion rate, and three-phase voltage imbalance in the associated local power grid. It triggers control warnings based on the voltage safety margin, the load safety margin, and the power quality margin, and controls the corresponding controlled devices according to the preset control strategy corresponding to the control warning.

6. The method according to claim 5, characterized in that, Also includes: The distribution area control center determines the voltage safety, load safety, and power quality safety based on the real-time node voltage and preset reference voltage, real-time branch power and rated power, real-time voltage harmonic distortion rate, and three-phase voltage imbalance in the associated distribution area power grid. The voltage safety margin, load safety margin, and power quality safety margin are weighted and summed to obtain the power grid safety margin of the distribution area. If the power grid safety margin of the distribution area continues to decrease within a first preset time period, the edge device control center is instructed to increase the acquisition frequency of real-time node voltage, real-time branch power and rated power, and real-time voltage harmonic distortion rate, as well as increase the determination frequency of the voltage safety margin, load safety margin, and power quality margin.

7. The method according to claim 1, characterized in that, Before the edge device control center outputs device control commands based on the received subset of commands using the second input convex neural network, the following steps are also included: If the edge device control center does not receive a subset of instructions from the area control center within a second preset time period, it uses an autoregressive integral moving average model to predict a subset of instructions based on historical instruction subsets. The edge device control center utilizes a second input convex neural network to output device control commands based on a subset of received commands, including: The edge device control center uses a second input convex neural network to output device control commands based on the instruction subset.

8. The method according to claim 1, characterized in that, The loss function of the first input convex neural network is determined in the following ways: The distribution area control center determines the grid economic factor using the real-time electricity price, purchased and sold electricity volume, and active power loss of the distribution area grid; it determines the voltage quality factor based on the real-time node voltage and preset voltage deviation threshold of the distribution area grid; it determines the power quality factor based on the real-time voltage harmonic distortion rate and real-time three-phase voltage imbalance of the distribution area grid; and it determines the safety constraint violation degree factor based on the target parameters of the distribution area grid, wherein the target parameters include real-time node voltage, real-time branch power, real-time branch current, real-time grid frequency, real-time apparent power of equipment, real-time state of charge of energy storage, and power change rate; and it performs a weighted summation of the grid economic factor, the voltage quality factor, the power quality factor, and the safety constraint violation degree factor to obtain the loss function of the first input convex neural network. The loss function of the second input convex neural network is determined in the following ways: The edge device control center determines the first squared value of the Euclidean distance between the device control command output by the second input convex neural network during training and the input quantity, wherein the input quantity is a subset of the commands output by the first input convex neural network after training; determines the second squared value of the Euclidean distance between the device control command output by the second input convex neural network during training and the local optimal command, wherein the local optimal command is a control command obtained after processing the historical local power grid state based on preset rules; and performs a weighted summation of the first squared value and the second squared value to obtain the loss function of the second input convex neural network.

9. A coordinated control system for transformer substations, characterized in that, The transformer area collaborative control system includes a transformer area control center and an edge device control center, wherein: The instruction distribution module in the distribution area control center is used to output a set of collaborative control instructions based on the global state information of the distribution area power grid using a first input convex neural network, and distribute a subset of instructions in the set of collaborative control instructions to the corresponding edge device control center. The collaborative control instructions in the subset of instructions are control parameters for each type of controlled device. The control module in the edge device control center is used to output device control commands based on a subset of received commands using a second input convex neural network, and send the device control commands to the corresponding controlled devices; determine the power grid safety margin based on the first local power grid status information associated with the current edge device control center, and control the corresponding controlled devices based on the power grid safety margin, wherein the power grid controlled by the distribution area control center includes the local power grid.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that are used to cause a processor to execute the area collaborative control method according to any one of claims 1-7.

Images